You are on page 1of 11

Mosaic structure in epitaxial thin films having large lattice mismatch

V. Srikant, J. S. Speck, and D. R. Clarke

Citation: Journal of Applied Physics 82, 4286 (1997); doi: 10.1063/1.366235


View online: http://dx.doi.org/10.1063/1.366235
View Table of Contents: http://scitation.aip.org/content/aip/journal/jap/82/9?ver=pdfcov
Published by the AIP Publishing

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
Mosaic structure in epitaxial thin films having large lattice mismatch
V. Srikant, J. S. Speck, and D. R. Clarkea)
Materials Department, College of Engineering, University of California-Santa Barbara, Santa Barbara,
California 93106-5050
~Received 10 April 1997; accepted for publication 24 July 1997!
Epitaxial films having a large lattice mismatch with their substrate invariably form a mosaic
structure of slightly misoriented sub-grains. The mosaic structure is usually characterized by its
x-ray rocking curve on a surface normal reflection but this is limited to the out-of-plane component
unless off-axis or transmission experiments are performed. A method is presented by which the
in-plane component of the mosaic misorientation can be determined from the rocking curves of
substrate normal and off-axis reflections. Results are presented for two crystallographically distinct
heteroepitaxial systems, ZnO, AlN, and GaN ~wurtzite crystal structure! on c-plane sapphire and
MgO ~rock salt crystal structure! on ~001! GaAs. The differences in the mosaic structure of these
films are attributed to the crystallographic nature of their lattice dislocations. © 1997 American
Institute of Physics. @S0021-8979~97!04821-4#

I. INTRODUCTION tially thicker than the absorption length of the x-rays thereby
precluding this direct measurement.
For many technological applications, it is necessary to In this article we describe an approach to determine the
grow thin films on substrates with large lattice mismatches or in-plane mosaic in heteroepitaxial systems from measure-
even substrates having a different crystal structure, for ex- ments of both the substrate normal and other, off-axis rock-
ample perovskite ferroelectrics on MgO,1,2 MgO on GaAs,3,4 ing curves, the advantage being that these rocking curves can
GaN,5–7 AlN,8 ZnO9,10 or LiNbO3, 11 on sapphire. Although be measured in a reflection geometry. As examples, analysis
epitaxy is generally achievable in these systems, the films of four systems, ZnO, AlN, and GaN on c-plane sapphire
usually exhibit a mosaic structure, meaning that the films and MgO on GaAs ~001!, are presented. First, we discuss our
consist of sub-grains slightly misoriented with respect to measurements and then propose an empirical model for de-
each other and the underlying substrate. In general the mo- termining the in-plane mosaic. Finally, the principal features
saic structure is undesirable as it can have deleterious effects of a physical description, based on crystallography of the
on the optical and electrical properties of the film. For ex- lattice dislocations of these materials, is presented providing
ample, ZnO films on sapphire have low angle grain bound- a basis for the conclusions of the empirical model.
aries, generated as a consequence of the mosaic structure,
II. EXPERIMENTAL DETAILS AND RESULTS
that act as scattering centers and decrease the mobility of
transport electrons. Similarly, low-angle grain boundaries The ZnO films on c-plane sapphire and the MgO films
lead to surface grooving which is a strong scattering center on ~001! GaAs investigated were grown by pulsed laser ab-
for light and can lead to optical losses, for instance in lation while the AlN and GaN films on c-plane sapphire
LiNbO3 waveguides grown on sapphire. were grown by metal-organic chemical vapor deposition
The mosaic of a nominally epitaxial film can be de- ~MOCVD!. The ZnO film was deposited in a background
scribed in terms of two crystallographic parameters: the oxygen pressure of 10 mTorr and a substrate temperature of
range of tilt that the sub-grains have with respect to the sub- 600 °C ~see Ref. 10 for more details! to a film thickness of
strate normal and their range of twist about the substrate 200 nm. The MgO film was grown at a substrate temperature
normal. The former quantity, usually expressed as the full- of 450 °C and a background pressure of 0.1 mTorr ~see Ref.
width half maximum ~FWHM! of the surface normal x-ray 4 for more details! to a film thickness of 200 nm. The GaN
rocking curve12 is often quoted in the literature and com- film was deposited at a substrate temperature 1080 °C to film
monly used as a parameter to evaluate the ‘‘quality’’ of the thickness of 2 mm and the AlN film was deposited at a sub-
film. However, a complete description of the mosaic struc- strate temperature of 1075 °C to a film thickness of 270 nm
in a Thomas Swan MOCVD reactor ~see Refs. 7 and 8 for
ture requires both the out-of-plane and the in-plane misori-
more details!. Using on-axis and off-axis x-ray diffraction
entations to be fully specified. Unfortunately, unlike the x-
scans, the epitaxial relationship between ZnO ~AlN, GaN!
ray rocking curves for the surface normal Bragg peaks,
and c-plane sapphire was determined to be ~0001! ZnO
which can be measured routinely for all samples with con-
~AlN, GaN!//~0001! Al2O3 and @101̄0# ZnO ~AlN, GaN!//
siderable accuracy, the in-plane rocking curve is less ame-
@112̄0# Al2O3 and that between MgO and GaAs was found to
nable to measurement. This is because in-plane x-ray rocking be ~001! MgO//~001! GaAs and @100# MgO//@100# GaAs.
curves must be measured in a transmission geometry. In the In order to evaluate the quality of the films, we measured
majority of practical cases the substrate is usually substan- the x-ray rocking curves of the surface normal and other
off-axis peaks ~see Tables I–IV!. The high resolution x-ray
a!
Electronic mail: clarke@engineering.ucsb.edu rocking curves ~v scans! were performed using a Philips

4286 J. Appl. Phys. 82 (9), 1 November 1997 0021-8979/97/82(9)/4286/10/$10.00 © 1997 American Institute of Physics
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
TABLE I. Details on the rocking curves on ZnO on c-plane sapphire.

Diffraction Angle of %
peak inclination FWHM Error Lorentzian Error

~0 0 2! 0 0.64 0.0065 58.1 3.8


~0 1 3! 32.43 0.91 0.0094 44.6 4.5
~0 1 2! 43.64 0.95 0.013 53.3 5.5
~0 1 1! 62.32 1.03 0.017 57.3 6.8

Materials Research Diffractometer equipped with a four


crystal monochromator, delivering a pure Cu K a 1 line of
wavelength, l50.15406 nm. In combination with the re-
ceiving slit mode a resolution of ;40 arcsec could be
achieved @as confirmed by the ~006! peak of the sapphire
substrate#. All the rocking curves were obtained in the sym-
metric geometry ( v 5 u ). For the off-axis scans this was
accomplished by tilting the substrate about the C axis by the
appropriate angle @see Fig. 3~a!#.
Figure 1 shows two representative rocking curves for the
film and the substrate ~in this case ZnO and sapphire, respec-
tively!. Figure 1~a! shows the measured rocking curve of the
~0002! ZnO and ~0006! sapphire diffraction peaks. Figure
1~b! shows the ~011̄2! ZnO and ~011̄2! sapphire peaks. Com-
pared to the ZnO film peak from the film the sapphire sub-
strate peak is extremely sharp. This indicates that the instru-
ment resolution about the v axis can be considered to be a
‘‘delta function’’ ~Even in the case of GaN films—which
have the smallest FWHM—this is a valid assumption as the
diffractometer resolution is one eighth of the smallest rock-
ing curve peak width measured on the GaN film!. In addi-
tion, in each of the four cases, broadening due to short co-
herence lengths can be neglected and all of the broadening FIG. 1. Comparison of the rocking curves of the ZnO film and the sapphire
substrate peaks for ~a! surface normal ~b! off-axis at an angle of 42.5°. The
observed in v scans can be attributed to mosaic spread. sharpness of the substrate peaks reflect the resolution of the diffractometer.
Broadening due to short coherence lengths can be estimate
from either asymmetric u 2D v scans13 or by determining
the smallest domain size via microscopy. In the materials
systems we have investigated broadening due to short coher- coherence length is small compared to the total measured
ence length is a concern only in the case of GaN films, as in spread.
all other cases the smallest domain sizes, as determined by In keeping with earlier observations,16,17 we have used a
structural characterization ~transmission electron micros- Psuedo–Voigt18 function to fit the rocking curves. Four pa-
copy, atomic force microscopy, and field emission scanning rameters define a Pseudo–Voigt function: the maximum in-
electron microscopy! is greater than 200 nm ~AlN,8 ZnO,14 tensity, the center ~mean! of the distribution, the FWHM, W,
and MgO!15 leading to a broadening of less than 0.05°, and the Lorentzian content, f , of the distribution ~see the
which is much smaller than the total measured spread Appendix!. Of these only the latter two are important for
~0.64°–1.0° for ZnO, 0.22°–0.96° for AlN, and 0.9°–1.3° distribution analysis and their values are listed in Tables
for MgO!. Using the data published in Ref. 5 it can be shown I–IV for the measured rocking curves on the ZnO, MgO,
that even in the case of GaN the broadening due to short AlN, and GaN films, respectively.
The results of the distribution analysis tabulated are also
represented graphically in Fig. 2. To determine the in-plane
TABLE II. Details on the rocking curves of MgO on GaAs. mosaic ~i.e., the W at an angle of inclination of 90°! from the
data in Fig. 2 it is necessary to know the functional depen-
Diffraction Angle of %
dence of the W of the reciprocal lattice ~RL! vectors as a
peak inclination FWHM Error Lorentzian Error
function of the angle of inclination. In the following section
~2 0 0! 0 0.88 0.004 33.5 1.5 an empirical approach to determine the above functionality is
~3 1 1! 25.23 0.9 0.0077 55 3.0 described and it is shown that there are fundamental differ-
~2 2 0! 45 1.17 0.005 46.5 1.5
~4 2 0! 26.56 0.90 0.0052 46.6 2.1
ences between the epitaxial materials studied and these dif-
~1 1 1! 54.47 1.30 0.014 85 3.2 ferences are related to their crystallographically allowable
dislocations.

J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke 4287
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
FIG. 2. FWHM as a function of inclination angle of the corresponding RL vector with respect to the surface normal for ~a! ZnO on c-plane sapphire, ~b! MgO
on GaAs, ~c! AlN on c-plane sapphire, and ~d! GaN on c-plane sapphire. Where the ‘‘error bars’’ are not visible, the error is less than or equal to the symbol
size. The curve labeled ‘‘independent’’ is fitted using Eqs. ~1!, ~2!, and ~3!, while the curve labeled ‘‘interdependent’’ is fitted using Eqs. ~1!–~5!.

III. ORIENTATION DISTRIBUTION ANALYSIS of representation, in its most general form a horizontal cross-
section of the distribution will reflect the anisotropy of the
To describe the crystallographic distribution of the sub-
film/substrate system. Since all of the material systems ana-
grains in an epitaxial film we make use of two reference
lyzed in this article show in-plane symmetry, the subsequent
systems. One is attached to the substrate, henceforth referred
analytical development has been carried out assuming this
to as the substrate reference frame ~SRF!. The other is at-
symmetry. We choose the CRF axes to coincide with RL
tached to each sub-grain, henceforth referred to as the crys-
vectors of the film and assume that their respective distribu-
tallite reference frame ~CRF! @see Fig. 3~a!#. Utilizing these
coordinate systems, the three-dimensional orientation distri- tions about the SRF can be measured. We then pose the
bution of the individual sub-grains can be mapped onto a following question, ‘‘can the orientation distribution of an
two-dimensional distribution of each of the CRF axes about arbitrary RL vector be described in terms of the measured
the SRF axes as shown schematically in Fig. 3~b!. Although distributions of the CRF axes?’’ @This is shown schemati-
the distribution has been drawn assuming mosaic symmetry cally in Fig. 3~c!#. If this question can be answered then, as
about the substrate normal ~‘‘in-plane’’ symmetry! for ease shown in this work, it is possible to deduce the in-plane
orientation distribution of an epitaxial film from measure-
ments made only in reflection geometry.
TABLE III. Details on the rocking curves on AlN on c-plane sapphire. To describe the distribution of any RL vector in the
plane defined by two CRF axes, in particular the surface
Diffraction Angle of %
normal ~tilt distribution! and an in-plane ~twist distribution!
peak inclination FWHM Error Lorentzian Error
RL vector as shown in Fig. 3~c!, two pieces of information
~0 0 2! 0 0.226 0.002 95 1 are needed. One is the component of each of the distributions
~0 1 3! 32.04 0.6 0.01 66 4 as a function of angle of inclination. The other is the manner
~0 1 2! 42.99 0.72 0.02 76 6
~1 1 2! 58.45 0.86 0.02 67 7
in which the two distributions combine. As stated earlier
~0 1 1! 61.84 0.96 0.02 58 4 there are two characteristics of the distribution of typical
rocking curves,19 their FWHM, W, and their Lorentzian frac-

4288 J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
FIG. 3. Schematic showing the relationship between the
thin film and the substrate in large lattice mismatched
system. ~a! The film is assumed to be made up of crys-
tallites with each having its own reference frame ~CRF!.
The relationship between the substrate reference frame
~SRF! and the diffractometer axes is also shown. ~b!
shows the orientation distribution of the sub-grains with
respect to the SRF. The measured sections are also
shown. ~c! The problem of describing ‘‘the distribution
of an arbitrary RL vector in terms of the tilt and twist
distributions’’ is posed graphically.

tion, f . From the data in Tables I–IV we can obtain a value angle! and the in-plane RL vector ~twist angle!, respectively.
of f , 20 for each of the material systems. In determining a Knowing the component of the distributions of the two CRF
component of the distribution of a CRF, we are primarily as a function of angle of inclination and the value of f from
concerned with how the W of the given CRF varies as a
function of angle of inclination. This is accomplished by
making use of rotation matrices as given by the theory of TABLE IV. Details on the rocking curves of GaN on c-plane sapphire.
rigid body rotations. Thus, Diffraction Angle of %
21 peak inclination FWHM Error Lorentzian Error
0 @ G 0 # 5cos @ cos ~ G 0 ! cos~ W y ! 1sin ~ G 0 !# ,
W tilt ~1!
2 2

21
~0 0 2! 0 0.083 1.3e-5 35 0.5
0 @ G 0 # 5cos @ sin ~ G 0 ! cos~ W z ! 1cos ~ G 0 !# ,
W twist ~2!
2 2
~0 1 3! 33.15 0.101 0.0003 42 0.9
~0 1 2! 44.34 0.111 0.00046 38 1.4
where G 0 is the angle of inclination ~i.e., the angle between ~1 1 2! 59.56 0.125 0.00037 37 1
the RL vector of interest and the surface normal! and W y and ~0 1 1! 61.58 0.127 0.0003 32 0.9
W z are the FWHM of the surface normal RL vector ~tilt

J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke 4289
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
Tables I–IV, we now need to determine the correct math- B. Inter-dependent distributions
ematical formulation for combining these two distributions.
In postulating the independence of the tilt and the twist
If the distributions are independent ~i.e., there is no interac-
distributions in the preceding sub-section, it was implicitly
tion between the tilt and the rotation of the sub-grains about
assumed that the individual sub-grains can rotate and tilt
the surface normal! then from the probability theory,21 the
about the substrate surface normal without affecting one an-
resultant distribution is simply the convolution of the two
other. This need not necessarily be so since the manner in
independent distributions. However, if the distributions are
which the rotation misfit of one sub-grain is accommodated
inter-dependent, then a conditional probability distribution
crystallographically might ‘‘restrict’’ or ‘‘enhance’’ the crys-
for one of the distributions with respect to the other must be
tallographic tilt with respect to the surface normal and vice-
defined and then convolution is carried out with respect to
versa. ~It is assumed that the sub-grains are strain free, that
this function. First, we consider the case where the distribu-
the individual grains completely tile the surface of the sub-
tions are assumed to be independent and later introduce a
strate, and no wedge-shaped cracks exist between grains!.
new parameter, m, that will account for the inter-dependence
Assuming this interaction to be exponential in nature, we
of the two distributions.
define a new effective W as follows:

S D
A. Independent distributions
0 @ G0#
W tilt
eff @ G 0 # 5W 0 @ G 0 # exp 2m
W twist ~4!
twist
Assuming that the tilt and the twist distributions are in- ,
W tilt
0 @0#
dependent, then the angular component of the two distribu-

S D
tions can be convoluted to give the resultant misorientation
0 @ G0#
W twist
eff@ G 0 # 5W 0 @ G 0 # exp
W tilt 2m ~5!
distribution. However, while analytical descriptions for the tilt
,
convolution of two Gaussian or two Lorentzian functions W twist
0 @ 90#
exist,21 no such analytical expression exists for the convolu-
where m is a new parameter characterizing the inter-
tion of two Pseudo–Voigt functions. In the absence of such a
dependence between the two distributions. W tilt 0 @ G0#,
convolution expression, we have developed an approxima-
W twist
0 @ 0# G are given by Eqs. ~1! and ~2!, respectively, and
tion ~detailed in the Appendix! that allows the W of the
W tilt
0 @ #0 and W twist
0 @ 90 # are the tilt and twist angles, respec-
resultant Pseudo–Voigt distribution to be expressed in terms
tively. The exponential dependence assumed here is purely
of the W of the initial Pseudo–Voigt functions. We have
phenomenological and has no precedent. It however has the
numerically determined that the following expression closely
basic characteristics that are required by our model. It is a
describes the W of the numerical convolution of two
monotonically increasing/decreasing function with param-
Pseudo–Voigt functions:
eters that can be used to control the rate of rise/decay. Al-
though we are unable to physically justify its use, it may
have its underpinnings in Boltzmann statistics. Furthermore,
W ~ resultant! 5 $ @ W ~ 1 !# n 1 @ W ~ 2 !# n % 1/n , ~3! the terms in the exponential have been normalized such that
m can only lie between 21 and 1. However, we find that in
all cases the value of m is positive. Thus, the physical sig-
where W(1) and W(2) are the FWHM of the initial distri- nificance of the interaction between the tilt and the twist
butions and n is a constant that varies between 1 and 2 de- distributions is to reduce the magnitude of the attainable tilt
pending on the value of f . It is important to note here that or twist below its maximum value, as defined by Eqs. ~1! and
the approximation used makes the assumption that the two ~2!, i.e., they are anti-correlated.
initial Pseudo–Voigt functions and the resultant Pseudo– Using the proposed effective W @Eqs. ~4! and ~5!# in
Voigt function all have the same value of f . conjunction with Eqs. ~1! and ~2! we fit Eq. ~3! to the data in
Using the above analysis and Eqs. ~1!–~3! the data in Figs. 2~a!–2~d!. In these cases, two fitting parameters are
Figs. 2~a!–2~d! are fit using the in-plane twist angle, W z , as used, W twist90
0 ~the twist angle! and m. The result of this fit is
a fitting parameter. These curves are labeled ‘‘independent’’ labeled ‘‘inter-dependent’’ in Fig. 2. As is evident, better fits
in the figures. Based on this fit, the in-plane mosaic is deter- are achieved in all cases. In the case of the ZnO, AlN, and
mined to be 0.94° for our ZnO film on sapphire, 0.93° for the GaN films the inter-dependence parameter, m ~50.07, 0.1,
AlN film on sapphire, 0.12° for the GaN film on sapphire, and 0.2, respectively! is small and the twist angle determined
and 1.0° for our MgO film on ~001! GaAs. It is evident from is not significantly different from the determination assuming
Fig. 2 that the fit for the ZnO and AlN on sapphire describes independent distributions presented in the previous sub-
the experimental data rather well. Although the fit for the section, 1.0° for the ZnO and AlN films as opposed to 0.94°
GaN film on sapphire does not describe the experimental and 0.93°, respectively, and 0.14° as opposed to 0.12° for the
data as well as that for ZnO and AlN, it nevertheless shows GaN film. However, in the case of the MgO thin films, the
the right trend. However, the fit for the MgO film on GaAs interaction parameter (m50.78) is large, indicating a strong
does not describe the experimental data very well. The poor inter-dependence between the tilt and the twist distributions.
fit in the case of MgO is attributed to our assumption that In consequence, the twist angle is also larger than that pre-
there is no inter-dependence between the tilt and the twist dicted on the basis of the earlier independent distribution
distributions. This conclusion is borne out by the analysis of model ~2.24° as opposed to 1.0°!. The only previous attempt
the next section where the effects of inter-dependence are that we know of to directly measure the in-plane twist in
explicitly treated. these systems has been by Ponce et al.22 for their GaN films

4290 J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
on c-plane sapphire using transmission electron microscopy
~TEM!. They quote a value of 0.133° for their twist angle
~their measured tilt angle is 0.083°!. This matches well with
the predictions of our empirical model.
The above analysis also indicates that there is a funda-
mental difference between the mosaic structure of an epitax-
ial ZnO, AlN, and GaN film on sapphire and an epitaxial
MgO film on GaAs. This difference is manifest in the mag-
nitude of the inter-dependence parameter m. The value of m
indicates that in the ZnO, AlN, and GaN films the tilt and the
twist distributions are independent while those in MgO films
are strongly inter-dependent. To provide further insight into
the empirical model and the differences in the mosaic struc-
ture predicted by the empirical model for the two cases con-
sidered, in the next section we develop a physical framework
based on dislocation geometry.

IV. DISCUSSION
It is evident from the preceding analysis that the corre- FIG. 4. Four different misfit dislocation configurations that can form at a
film substrate interface are shown. Of these configurations ~a! and ~c! lead to
lation between the tilt and the twist of the sub-grains is spe-
misaligned sub-grains while configurations ~b! and ~d! result in aligned
cific to the particular film and film/substrate combination. sub-grains.
While in ZnO, AlN, and GaN, the distributions can be mod-
eled as independent ~or nearly independent! of each other, in
MgO on ~001! GaAs they are seen to be strongly inter- misfit segment24 of the dislocations at the interface between
dependent. This inter-dependence was quantified in the pre- the film and substrate, configuration ~a! in Fig. 4 would lead
vious section by the introduction of an ad hoc inter- to tilt misorientation between the sub-grain and the substrate
dependence parameter, m. Here, we introduce a conceptual whereas configuration ~c! will lead to rotational misorienta-
description of the mosaic structure that is based on the type tion. In contrast configurations ~b! and ~d! in Fig. 4 will not
of dislocations necessary to accommodate both the misorien- lead to any misorientation between the sub-grain and sub-
tation between the sub-grains and the substrate and the mis- strate. The threading segment of the dislocations will simi-
orientation between the sub-grains themselves.23 The actual larly play an important role in the mosaic structure forma-
configuration of dislocations that will form cannot be ad- tion.
dressed in this manner but it does describe those that are In the following we primarily concern ourselves with
geometrically necessary to fill space and account for the sub- dislocation networks that lead to a misorientation between
grain misorientation distribution. Henceforth we refer to the film and the substrate. Of these, we will concern our-
these dislocations as ‘‘geometrically necessary disloca- selves with only the simplest configurations like those in
tions.’’ A key, but unjustified assumption is made, namely Figs. 4~a! and 4~c!. This then allows us to look at the orien-
that the dislocations necessary to describe the misorienta- tation of a single dislocation line and its Burgers vector and
tions can only be allowed lattice dislocations in the film. This predict the resulting tilt or twist of the sub-grain formed by a
neglects the possibility that the dislocations occurring at the collection of such dislocations. In the following paragraphs
film/substrate interface may have a crystallographic nature we discuss the different dislocation geometries that are pos-
different from that of lattice dislocations. However, as will sible in ZnO, AlN, GaN, and MgO based on their principal
be seen, the ‘‘geometrically necessary dislocation’’ frame- slip systems and the epitaxial relationship between the film
work provides a conceptually straightforward way of de- and substrate. We show that the two in combination are con-
scribing the correlation between the twist and tilt misorien- sistent with the different magnitudes of interaction deter-
tations of the sub-grains. There will, in addition, be a mined in the previous section.
population of randomly distributed dislocations. Such dislo-
A. ZnO, AlN and GaN on c -plane sapphire
cations do not produce any net rotations or twists nor do they
relieve any misfit strain but may, of course, have adverse ZnO, AlN, and GaN have a wurtzite structure and grow
effects on the electrical and optical properties of the films. A with their c axis oriented parallel to the surface normal on
fuller mathematical description of the geometrically neces- c-plane sapphire. The primary slip systems of ZnO, AlN, and
sary dislocations will be presented in a subsequent publica- GaN are known to be $0001% ^12̄10& and $101̄0% ^12̄10&.25 In
tion. Fig. 5 three different dislocation slip systems ~D1, D2, and
Intuitively, the difference in the degree of independence D3! are shown. For the slip system indicated D1,26 there are
must be related to the growth mechanisms underlying the three possible dislocation line/Burger’s vector configuration
formation of mosaic structure. At the dislocation level, the possible. Of these, only one is depicted in Fig. 5. In this
crystallographic mosaic structure can be described in terms particular configuration the misfit dislocations are pure screw
of geometrically necessary dislocations that accommodate dislocations and lead only to twist between the sub-grains
the mosaic structure. For example, by consideration of the and the substrate. The other two Burgers vectors ~at 120° to

J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke 4291
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
FIG. 5. Shows the various active slip systems in a Wurtzite crystal structure. The specific geometry is determined by the epitaxial relationship between the
film and the substrate.

the one shown in the figure! would lead to a dislocation those that lead to tilt (D3) of the sub-grains with respect to
having a mixed character. The screw component, as in the the substrate in these films, are independent. This is consis-
previous case, leads to a twist between the film and the sub- tent with our observations in the previous section, where it
strate. However, the edge component, which relieves lattice was concluded that there is a very small inter-dependence
parameter mismatch strain, plays no role in the misorienta- between the twist and tilt distribution of the sub-grains in
tion between the sub-grain and the substrate. For the slip ZnO, AlN, and GaN thin films on c-plane sapphire.
system indicated (D2) in Fig. 5 the misfit segment is a pure
screw dislocation and can only lead to twist between the
B. MgO on GaAs substrates
sub-grain and the substrate. On the other hand the threading
segment plays no role in the misorientation between the sub- MgO grows with a cube-on-cube orientation on ~001!
grain and the substrate. The only way to achieve tilt between GaAs single crystal. MgO has a rock-salt crystal structure
the sub-grains and the substrate is to have a dislocation in the with $110% ^110& as the primary slip system. At high tem-
basal plane with its Burger’s vector out of the basal plane peratures ~well above 1000 °C!32 a secondary slip system,
~see dislocation slip system labeled D3 in Fig. 5!. The $100% ^110&, also operates. As the growth temperature is far
$101̄0% @0001# dislocation slip system has been observed in below that needed to cause slip on this secondary slip sys-
bulk BeO,27 a crystal with the same wurtzite crystal struc- tem, we assume, for the present analysis, that slip only oc-
ture, at high stresses ~230 MPa!. Such dislocations have also curs on the primary slip systems. The primary slip systems
been observed in both homoepitaxial28 and hetero- are shown in Fig. 6. In this dislocation geometry there are six
epitaxial29,30 GaN films.31 As far as we know, no such de- independent slip systems, one group of four and another
tailed studies are available for ZnO or AlN films on sapphire. group of two. The dislocation slip system indicated as D1 is
However, we assume that such lattice dislocation slip sys- one slip system out of the group of four. We see that the
tems are operative in ZnO and AlN because of their similar- misfit segment of the dislocation leads to tilt, whereas the
ity to GaN and BeO. Dislocations on this slip system can threading segment of the dislocation leads to twist between
result in a tilt between the sub-grains and the substrate. The the sub-grains and the substrate. This indicates that by intro-
pure edge misfit segment of the dislocation leads to tilt be- ducing a tilt we are required by the dislocation geometry to
tween the sub-grain and the substrate, whereas the pure also introduce twist. Now, consider the slip system indicated
screw threading segment plays no role. The above discussion D2 in Fig. 6. Here the pure screw misfit segment leads to
suggests that the dislocations on the three slip systems twist between the substrate and the sub-grains, whereas the
~shown in Fig. 5!, those that lead to twist ~D1 and D2! and pure edge threading segment plays no role in the misorien-

4292 J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
FIG. 6. Shows the various active slip systems in a rock-salt crystal structure. The specific geometry is determined by the epitaxial relationship between the
film and the substrate.

tation between the sub-grain and the substrate. This indicates GaAs. The difference is attributed to the crystallographically
that there is no influence of introducing twist, through this allowed lattice dislocations required to geometrically accom-
slip system, on the tilt exhibited by the sub-grains. Thus it modate the misorientation.
can be said that four out of the six primary slip systems in
MgO can lead to both tilt and twist between the substrate and
the film. This is consistent with the strong inter-dependence ACKNOWLEDGMENTS
between the tilt and the twist distribution that was deter-
The authors would like to thank Dr. E. J. Tarsa for pro-
mined in the previous section from the rocking curve analy-
viding MgO/GaAs thin films and the MOCVD group at Uni-
sis.
versity of California at Santa Barbara for the GaN and AlN
The geometrically necessary dislocation framework
films on sapphire. This work was begun with supported by
seems to provide the essential tools to describe the mosaic
the Department of Energy, Office of Basic Energy Sciences,
structures in large lattice mismatched systems. Although no
under Grant No. DE-FG03-91ER45447 and made use of the
attempt has been made to address issues concerning the
Materials Research Laboratory’s Central Facilities supported
mechanisms of mosaic structure formation, the description
by the National Science Foundation under Award No. DMR-
does establish crystallographic constraints that should be fol-
9123048.
lowed during the growth of large mismatch films. Consider-
ation of geometrically necessary dislocations also provides a
physical basis for the parameters introduced in the empirical
APPENDIX
model discussed in the previous section.
Convolution of Pseudo–Voigt functions
V. SUMMARY No analytical solution exists for the convolution
Pseudo–Voigt ~PV! functions and consequentially it must be
A phenomenological model is presented that enables the
carried out numerically. In order to avoid the cumbersome
in-plane misorientation distribution of a mosaic structure to
nature of such numerical calculations, we have developed an
be obtained from a series of x-ray rocking curves measured
approximation to obtain the FWHM (W) of the resultant
in the reflection geometry. From the analysis, the degree of
distributions, for any Pseudo–Voigt function, from the W of
correlation can be established between the twist and tilt com-
the starting distributions. This approximation is based on the
ponents of the sub-grain misorientation distributions. The
following reasoning. A Pseudo–Voigt function is defined as
misorientation distributions are quite different for ZnO, AlN,
and GaN films grown on sapphire and MgO films grown on PV ~ x ! 5 ~ 12 f ! G ~ x ! 1 f L ~ x ! , ~A1!

J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke 4293
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
FIG. 8. Exponent, n, as a function of f , fraction of Lorentzian in the
Pseudo–Voigt function. The solid line is based on the equation in the inset
and shows very good correspondence to the numerical data.
FIG. 7. The FWHM of the convoluted distribution as a function of the
FWHM of one of the starting distribution is shown. The other distribution
had a constant width of 2. The solid lines are fits based on the equation
shown in the inset. The starting distributions and the convoluted distribu- a function of starting W as shown in Fig. 7. A value of 2 was
tions are forced to have the same fraction of Lorentzian character. chosen for the fixed W in all of convolutions. The points are
then fit to the function in Eq. ~A4! while making the substi-
tution, W(2)52. The resulting value of the exponent, n, is
where f is the fraction of Lorentzian character of the distri- then plotted as a function of f in Fig. 8. It is found that the
bution and G(x) and L(x) are Gaussian and Lorentzian points can be fit to a rather simple expression
functions defined as

S D
n511 ~ 12 f ! 2 . ~A5!
~ x2x 0 ! 2
G ~ x ! 5H * exp 2 4 ln~ 2 ! , ~A2! In using this expression, the rocking curves are first fit to a
W2

S D
Pseudo–Voigt function and then the value of f from the fit is
H 1 used to obtain the required exponent, n, from the above
L~ x !5 , ~A3!
@ p ln~ 2 !# 1/2 ~ x2x 0 ! 2 equation.
4 11
W2
1
H. A. Lu, L. A. Wills, and B. W. Wessels, Appl. Phys. Lett. 64, 2973
respectively. In these equations W is the FWHM and x 0 is ~1994!.
the mean of the distributions. In writing the above definitions 2
V. Srikant, E. J. Tarsa, D. R. Clarke, and J. S. Speck, J. Appl. Phys. 77,
of Gaussian and Lorentzian functions we have required that 1517 ~1995!.
3
the two distributions have the same W and the same area K. Nashimoto, D. K. Fork, and T. H. Geballe, Appl. Phys. Lett. 60, 1199
~1992!.
under the curve. 4
E. J. Tarsa, M. De Graaf, D. R. Clarke, A. C. Gossard, and J. S. Speck, J.
From standard probability theory, we know that analyti- Appl. Phys. 73, 3276 ~1993!.
5
cal solution for the convolution for Gaussian and Lorentzian B. Heying, X. H. Wu, S. Keller, Y. Li, D. Kapolnek, B. Keller, S. P.
distributions exist and are given by W(resultant) DenBaars, and J. S. Speck, Appl. Phys. Lett. 68, 643 ~1996!.
6
R. F. Davis, T. W. Weeks, M. D. Bremser, and S. Tanaka, Solid State
5 $ @ W(1) # 2 1 @ W(2) # 2 % 1/2 and W(resultant)5 @ W(1) # Commun. 41, 129 ~1997!.
1 @ W(2) # , respectively, where W(1) and W(2) are the 7
D. Kapolnek, X. H. Wu, B. Heying, S. Keller, B. Keller, U. K. Mishra, S.
FWHM of the starting distributions. Hence, we postulate P. DenBaars, and J. S. Speck, Appl. Phys. Lett. 67, 1541 ~1995!.
that, as the Pseudo–Voigt function varies from being 100%
8
A. Abare, V. Srikant, S. P. DenBaars, and D. R. Clarke ~unpublished!.
9
M. Kasuga and M. Mochizuki, J. Cryst. Growth 54, 185 ~1981!.
Gaussian to 100% Lorentzian, it should be possible to de- 10
V. Srikant, V. Sergo, and D. R. Clarke, J. Am. Ceram. Soc. 78, 1931
scribe the W of the resultant Pseudo–Voigt function in terms ~1995!.
11
of the W of the two starting functions as follows: D. K. Fork, F. Armani-Leplingrad, J. J. Kingston, and G. B. Anderson, in
Thin Film for Integrated Optics, Mater. Res. Soc. Symp. Proc. edited by
W ~ resultant! 5 $ @ W ~ 1 !# n 1 @ W ~ 2 !# n % 1/n , ~A4! B. W. Wessels and D. M. Walba ~MRS, Pittsburgh, PA, 1995!, p. 189.
12
The rocking curves are measured by keeping a constant angle between the
where n takes on values from 1 to 2 depending on the value source and the detector ~equal to twice the Bragg angle for the diffracting
of f in Eq. ~A3!. In order to test this postulate the following 13
planes! and then rocking the substrate about the surface normal.
numerical analysis was performed: A function of known P. F. Fewster, X-ray and Neutron Diffraction: Theory and Applications,
edited by A. Authier, S. Lagomarsino, and B. K. Tanner ~Plenum, New
value of f and a fixed W is convoluted by a distribution of York, 1996!, p. 269.
varying W ~from 0.5 to 4! with the same value of f . This 14
V. Srikant, PhD thesis, University of California, Santa Barbara, 1995.
convolution is performed numerically and the resultant dis-
15
X. H. Wu, E. J. Tarsa, and J. S. Speck ~unpublished!.
16
E. Galdecka, Acta Crystallogr., Sect. A: Found. Crystallogr. 49, 106
tribution is fit to a Pseudo–Voigt function having the same
~1993!.
value of f as the starting distributions. The calculated result- 17
E. Galdecka, Acta Crystallogr., Sect. A: Found. Crystallogr. 49, 116
ant W for the various starting distributions is then plotted as ~1993!.

4294 J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51
18
A Pseudo–Voigt function is a linear combination of a Gaussian and a 25
F. R. N. Nabarro, Theory of Crystal Dislocation ~Oxford University Press,
Lorentzian function. For a more detailed description see Appendix. Oxford, 1967!.
19 26
The rocking curve measures a vertical slice of the two-dimensional distri- The dislocations are expected to end at the sub-grain boundaries, edge of
bution function shown schematically in Fig. 3~b!. the film or react with dislocations from the same slip system or other slip
20
The value of f used is the average of all of the rocking curves measured. systems.
21
R. M. Sheldon, Introduction to Probability Models, 5th ed. ~Academic,
27
G. G. Bentle and K. T. Miller, J. Appl. Phys. 38, 4248 ~1967!.
28
F. A. Ponce, D. Cherns, W. T. Young, and J. W. Steeds, Appl. Phys. Lett.
Boston, 1993!.
69, 770 ~1996!.
22
F. A. Ponce, MRS Bull. 22, 51 ~1997!. 29
23 X. J. Ning, F. R. Chien, P. Pirouz, J. W. Yang, and M. Asif Khan, J.
Some of these dislocations may account for the misfit strain between the
Mater. Res. 11, 580 ~1996!.
subgrain and the substrate but they do not have to. For example, pure 30
X. H. Wu, L. M. Brown, D. Kapolnek, S. Keller, B. Keller, S. P. Den-
screw dislocations at the sub-grain/substrate interface, as depicted in Fig. Baars, and J. S. Speck, J. Appl. Phys. 80, 3228 ~1996!.
4~c!, do not account for any lattice parameter misfit between the sub-grain 31
In addition, threading dislocations having a burgers vector lying in the
and the substrate. @112̄3# direction, resulting from the interaction of dislocations having bur-
24
In this article we make use of the following terminology for convenience. gers vectors @0001# and @112̄0#, have also been reported. However, it can
The part of any dislocation lying at the sub-grain/substrate interface is be shown that such threading dislocations play no role in determining the
referred to as the ‘‘misfit segment’’ whether it actually takes part in re- misorientation between the film and the substrate.
lieving the misfit strain between the film and the substrate or not. Simi- 32
T. Vasilos and E. M. Passmore, in Ceramic Microstructures, Their Analy-
larly, the part of any dislocation threading through the film is referred to as sis, Significance and Production, edited by R. M. Fulrath and J. A. Pask
a ‘‘threading segment.’’ ~1966!, Chap. 17.

J. Appl. Phys., Vol. 82, No. 9, 1 November 1997 Srikant, Speck, and Clarke 4295
[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:
203.200.35.12 On: Mon, 17 Mar 2014 12:02:51

You might also like